0
$\begingroup$

I work on data from a cohort study (n = 3100). We want to compare outcomes of individuals with a special condition (n = 300) and their matched controls. I estimated propensity scores and used genetic matching in order to create a 2:1-matched data set. Because baseline covariates are quite imbalanced, I turned to matching with replacement, which led to much better post-matching balance of covariates.

Now, I want to perform survival-analysis using the Kaplan-Meier-Estimator. My question is: How do I account for the fact, that multiple control subjects are used several times in the control group? Simply put, can I draw "normal" KM-plots using all the "duplicates" in the control group? I know one has to factor in the weights assigned to these control subjects when calculating treatment effects. What I don't understand, however, how one is supposed to account for these dependencies when performing survival analysis.

$\endgroup$
0
$\begingroup$

As you correctly identify, one needs to account for both the pairing and the repeated use of the control units to validly account for all sources of dependence. Austin & Cafri (2020) explain how to do so and provide R code for their variance estimator.

(Note: the R code is in the supplementary files. After downloading the file, change the file ending to .pdf to open the file)

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.